7 SEO Workflows You Can Run From Cursor With MCP

Most “AI SEO” demos start the same way: paste a keyword list into a chat, ask for titles, and hope something ranks.

That can produce words. It rarely produces decisions.

The useful shift is smaller and more practical. Keep your editor open. Keep your site evidence close. Ask for work that starts from what Search Console and your crawl already know—not from a model’s guess about what “usually works.”

That is what SEO Perception MCP is for. On a paid plan, you create a personal access token under Account → MCP, connect your MCP client (Cursor is a common choice; Claude and other clients work too), and call tools that read stored weekly Search Console facts and crawl-backed possibilities. MCP does not call Google live. It does not refresh GSC, recrawl pages, or mutate issues. Setup detail lives in SEO Perception MCP docs.

Below are seven workflows you can run from Cursor once a site is connected. Each one maps to a real tool, a concrete prompt, and a decision outcome you can ship.

Start every session the same way: list your sites, pick a siteId, then go deep.

Before the seven: get oriented

Tool: seoperception_list_sites → seoperception_get_site_overview

Prompt:

List my SEO Perception sites. Then give me a bounded overview for [site]: top queries, traffic-leading pages, and rank momentum. Tell me what looks worth investigating this week and what can wait. Do not invent numbers that are not in the tool results.

Decision outcome: A short “what matters this week” note. Not a 40-page audit. Enough context that the next six workflows do not start from vibes.

If the overview is empty, stop. Confirm Search Console collection and crawl finished in the product UI before you ask an agent to optimize anything.

---

1. Diagnose one page before you rewrite it

Tool: seoperception_get_page_diagnostics

Rewrites fail when nobody knows whether the problem is content, title/snippet fit, indexability, competing URLs, or simply weak demand.

Prompt:

Diagnose https://example.com/your-page/ for this site. Combine latest crawl metadata, stored GSC weekly trend, top queries, and open opportunities. Separate: (1) technical blockers, (2) query/content mismatch, (3) title or snippet problems, (4) work that can wait. Recommend the smallest first fix.

Decision outcome: A page brief with a primary fix type. Example decisions: “update H1 and first section for query cluster X,” “fix canonical/indexability first,” or “do not rewrite—support with internal links instead.”

This is the workflow to run before you spend a morning “improving” a page that was never crawlable in the first place.

---

2. Find losing queries on a page that still has impressions

Tool: seoperception_search_queries (optionally with page diagnostics)

A page can still show up and still lose. Impressions hold. Clicks fall. Position drifts. Or the page ranks for a cluster it barely mentions.

Prompt:

Search stored queries for this site, sorted by clicks descending, hide branded where useful. Focus on queries landing on /your-page/. Flag queries with meaningful impressions but weak CTR or soft position. For each, say whether the fix is title/meta, on-page answer, or intent mismatch. Suggest concrete copy changes tied to the query text.

Decision outcome: A short recovery list: which queries to reclaim on this URL, which to ignore as noise, and which to move to a better URL.

Pair this with the in-app prioritization patterns in SEO Possibilities when you want the same evidence outside the editor.

---

3. Surface content gaps that keyword lists miss

Tool: seoperception_find_content_gaps

Content gaps in SEO Perception are not “competitors rank for X and you do not.” They are stored evidence patterns: query not in content, low CTR, snippet/title alignment, question queries, refresh candidates.

Prompt:

Find content gaps for this site. Group them into: update existing page, add a section, create a new supporting page, or ignore. Prefer non-branded question queries and query-not-in-content cases. For each keep: query, landing URL, why it is a gap, and the smallest evidence-backed action.

Decision outcome: A content queue with update-first bias. New URLs only when no existing page can own the intent cleanly.

If you want the strategic why behind “gaps bigger than keyword lists,” read Query fan-out means your content gaps are bigger than your keyword list. MCP is how you operationalize that idea on your own data.

---

4. Align titles and snippets to the queries you already earn

Tool: seoperception_find_content_gaps + seoperception_get_page_diagnostics

Low CTR with stable impressions is often a packaging problem: title, meta description, H1, and opening paragraph do not match the query language Google already associates with the URL.

Prompt:

From content gaps and page diagnostics, list pages with low-CTR or snippet/title alignment issues. For each page, propose one title and one meta description rewrite that reflects the top stored queries without keyword stuffing. Call out when the page content itself cannot support the promised title.

Decision outcome: A safe title/meta batch for pages that already have demand. No new article required.

This is one of the highest leverage “small edits” workflows because it uses queries you already earn, not invented primary keywords.

---

5. Resolve keyword cannibalization with a preferred page

Tool: seoperception_find_cannibalization

When multiple URLs earn the same query, Google still has to pick. Your site often splits attention instead of clarifying ownership. SEO Perception persists multi-page query conflicts with preferred and competing URLs so you can consolidate or differentiate on purpose. Product detail: Keyword cannibalization across URLs.

Prompt:

Find keyword cannibalization for this site. For each conflict: preferred URL, competing URLs, and a recommendation—consolidate, differentiate intent, redirect, or leave alone. Explain the recommendation using the stored evidence, not generic SEO theory.

Decision outcome: A conflict board where every row has an owner URL and a next action.

Do not “add the keyword to every page.” Pick a preferred page, then use links and differentiation so the rest of the site supports that choice. Deeper walkthrough: Find keyword cannibalization with an AI agent and MCP.

---

6. Strengthen internal links where ranking potential already exists

Tool: seoperception_suggest_internal_links

Internal links are how your site explains relationships. Underlinked pages with real query evidence are a structure problem, not a “need more content” problem. See also Internal links are how your site explains itself.

Prompt:

Suggest internal links for this site. Prioritize destinations with observed ranking potential. For each suggestion: source context, destination, proposed anchor, and why the link helps the reader. Reject any suggestion that looks like exact-match spam or orphan-page stuffing.

Decision outcome: A small link plan you can implement in a day: five to fifteen high-quality links, not a thousand automated insertions.

Use this after cannibalization decisions so you reinforce the preferred URL instead of feeding competitors on your own domain.

---

7. Prioritize open opportunities by severity and recency

Tool: seoperception_list_opportunities

When everything is “important,” nothing ships. Opportunity listing returns bounded stored crawl and Search Console possibilities ordered by severity and recency—so an agent can build a week plan instead of a guilt list.

Prompt:

List open SEO opportunities for this site, highest severity first. Summarize the top items as a weekly plan: do now, do this week, park. For each, state the evidence type (crawl vs GSC) and the first human action. Do not expand into a full rewrite for low-severity items.

Decision outcome: A Monday brief. Compatible with how Today’s Focus and SEO Possibilities frame prioritization in the product.

This workflow is also the clean handoff to a teammate: paste the plan into a ticket, keep the siteId and URLs attached, and leave the model chat behind.

---

How the seven fit together in one Cursor session

A practical session order:

1. List sites and overview.
2. List opportunities (week plan).
3. Pull cannibalization conflicts that would waste rewrite time.
4. Diagnose the top one or two URLs.
5. Search queries / content gaps for those URLs.
6. Suggest internal links to support the preferred pages.
7. Only then ask the agent to draft copy, titles, or FAQs.

That order matters. If you ask for a rewrite first, the model will happily invent structure. If you ask for evidence first, the rewrite has constraints.

Remember the privacy boundary: tool results are stored site analytics. Your MCP client may send them to a model provider. Treat them like you would treat an exported Search Console spreadsheet.

What MCP is not

MCP is not a live Google API proxy. It will not:

• fetch fresh Search Console data on demand
• run URL Inspection
• submit sitemaps
• trigger recrawls
• close or mutate issues for you

Those limits are intentional. The agent reasons over evidence SEO Perception already collected. You stay in control of collection cadence, crawl scope, and publishing.

For client setup, token limits, and the live tool catalog, use MCP integration and MCP for Cursor docs. Create and revoke tokens under Account → MCP.

The point

The win is not “AI wrote more SEO content.”

The win is: same editor where you change the site, same evidence you already trust, seven workflows that end in decisions—diagnose, recover, fill gaps, fix packaging, resolve conflicts, connect pages, prioritize.

If you can only run two this week, run opportunities and page diagnostics. Everything else gets sharper once those two are honest.

---

For related reading, continue with Turn Google Search Console queries into content briefs with MCP, Find keyword cannibalization with an AI agent and MCP, and SEO Possibilities.

Frequently asked questions

Does SEO Perception MCP call Google Search Console live?

No. Tools read stored weekly Search Console facts and crawl evidence already collected in SEO Perception. MCP does not call Google live, refresh GSC, or run URL Inspection.

Do I need a paid plan for MCP?

Yes. MCP personal access tokens and POST /mcp require a paid plan (Starter, Pro, Business, or Enterprise). Manage tokens under Account → MCP.

Can I use these workflows outside Cursor?

Yes. Any MCP client that can call the SEO Perception tools works. Cursor is a convenient place because it sits next to the code and content you are changing. Setup notes are in MCP for Cursor; the feature overview is on MCP integration.

What if a tool returns an empty list?

Usually the site has no matching stored evidence yet, or filters excluded everything. Confirm Search Console import and crawl freshness for that site, then retry with a broader prompt.

Should the AI rewrite pages automatically from MCP results?

Use MCP for diagnosis and briefs. Keep humans (or a careful review pass) on brand voice, legal claims, and publish decisions. Treat agent drafts as proposals grounded in evidence, not auto-shipped changes.

How is this different from pasting a GSC CSV into chat?

CSV paste is a one-off dump. MCP keeps a stable tool interface over the same stored facts and crawl-backed possibilities, with site scoping, pagination limits, and provenance that marks stored-data reads. You can repeat workflows without rebuilding the export each time.

---